fix: preserve transcription item id

[nightcityblade]

Fixes #6109

Summary

• expose item_id on UserInputTranscribedEvent
• pass through realtime input transcription item IDs when re-emitting user transcription events
• add a regression test for item ID propagation

Testing

• uv run pytest tests/test_agent_session.py -q
• uv run pytest tests/test_agent_session.py::test_realtime_user_input_transcription_preserves_item_id -q
• uv run ruff check livekit-agents/livekit/agents/voice/events.py livekit-agents/livekit/agents/voice/agent_activity.py tests/test_agent_session.py

[CLAassistant]

All committers have signed the CLA.

[devin-ai-integration]

Devin Review found 0 potential issues.

[areebkhan-tech]

it seems to have merge conflicts

[nightcityblade]

Thanks for flagging it — I'm working through a rebase now. The branch drift turned out to be larger than expected, so I want to untangle it cleanly before I push an update.

[nightcityblade]

Updated the branch to resolve the merge conflicts by merging the latest livekit/agents:main into fix/issue-6109 (no force-push).

Validation run:

• uv run pytest tests/test_agent_session.py::test_realtime_user_input_transcription_preserves_item_id ✅

The PR now reports as mergeable on GitHub. I also noticed the CLA check is still pending/not signed, which appears to be the remaining non-code item.

[devin-ai-integration]

Devin Review found 0 new potential issues.

[areebkhan-tech]

Hi, can u pls request re-review this has not been merged yet, or could it be you have not signed CLA yet .

[Monkiia]

I checked the branch again. The code change itself is already rebased and the targeted test passed in the last update; the remaining blocker appears to be the CLA status rather than an outstanding code review issue.

Once the CLA is recognized on GitHub, I'm happy to request re-review immediately.

[davidzhao]

@nightcityblade could you please sign the CLA?

[devin-ai-integration]

Devin Review found 0 new potential issues.

[nightcityblade]

I’m checking the CLA flow from this account now. The branch itself is already up to date and the targeted regression test passed in the last code update; once GitHub recognizes the CLA, I’ll request re-review immediately.

[theomonnom]

Pretty sure we will need to expose even more fields in the future, extra seems more appropriate

Suggested change

	item_id: str | None = None
	extra: dict = {}

We can then do something like extra["openai"]["item_id"] ?